Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
Improving the shape of the cross-correlation function for leak detection
in a plastic water distribution pipe using acoustic signals
Yan Gaoa*, Michael J. Brennanb, Yuyou Liuc,d, Fabrício C.L. Almeidae, Phillip F. Josephf
aKey Laboratory of Noise and Vibration Research, Institute of Acoustics, Chinese Academy of
Sciences, Beijing 100190, China
bDepartment of Mechanical Eng., State University of São Paulo (UNESP), Ilha Solteira
Campus, Av. Brasil Centro, 56, 15385-000, Ilha Solteira, Brazil
cBeijing Municipal Institute of Labour Protection, Beijing 100054, China
dAECOM Infrastructure & Environment UK Limited, London SW19 4DR, UK
eDepartment of Biosystem Eng., State University of São Paulo (UNESP), Tupã Campus, Av.
Rua Domingos da Costa Lopes, Jardim Itaipu, 780, 17602-496, Tupã, Brazil
fInstitute of Sound and Vibration Research, University of Southampton, Southampton SO17 1BJ,
UK
ABSTRACT
This paper is concerned with time delay estimation for the detection of leaks in buried plastic
water pipes using the cross-correlation of leak noise signals. In some circumstances the
bandwidth of over which the signal analysis can be conducted is severely restricted because of
resonances in the pipe system, which manifest themselves as peaks in the modulus of the power
spectral and cross spectral densities, and deviations from straight-line behaviour in the phase
of the cross spectral density. The result can be a cross-correlation function in which it is
difficult to estimate the time delay accurately. This paper describes a procedure in which the
shape of the cross-correlation function can be significantly improved, resulting in an
unambiguous and clear estimate of the time delay. The frequency response function(s) of the
resonator(s) responsible for the resonance effects are first determined and then the data is
processed using the model(s) of the resonators to remove these effects. This enables more
2
signal processing to be conducted, potentially over a much wider bandwidth, further improving
the shape of the cross-correlation function. The process is illustrated in this paper using
hydrophone measured data at a leak detection facility. The current limitation in the process is
that it is carried out manually, which could potentially restrict its application in practical
acoustic correlators. The challenge now is to develop an algorithm to carry out the procedure
automatically.
keywords: leak detection; water pipe; cross-correlation function; time delay estimation;
resonance.
*Corresponding author. E-mail address: [email protected]
3
1. Introduction
Water distribution networks are of paramount importance for maintaining a substantive
modern life and economic growth. Underground pipes are susceptible to leakage, due to
excavation damage, sabotage, deterioration and aging. Water leakage is a subject of increasing
concern across the world because of the potential danger to public health, economic constraints,
environmental damage and wastage of energy. Acoustic based leak detection techniques have
been developed over the past 30 years, and are in common use in water distribution networks
[1, 2]. One such technique uses cross-correlation of measured leak noise signals to determine
the difference in arrival times (time delay) between acoustic/vibration signals measured either
side of a water leak. Together with the knowledge of the speed of noise propagation, this is
then used to determine the location of the leak.
Time delay estimation (TDE) is of great interest in many engineering fields, such as
direction finding, source localization, and velocity tracking [3]. It can be carried out in the time
and/or frequency domains [4, 5]. Most commercial leak noise correlators utilize the basic cross-
correlation (BCC) function via the Fast Fourier transform for TDE. In plastic pipes, measured
water leak noise mostly occurs at low frequencies, below about 200 Hz, although this can be
higher for leaks close to a measurement point. The reason for this predominantly low frequency
content, is that a plastic pipe essentially acts as an acoustic low-pass filter, which degrades the
TDE procedure using the conventional BCC, in particular in the case of a poor signal-to-noise
ratio [6, 7]. To overcome this problem, Gao et al. [4] applied a pre-whitening process, using
appropriate frequency weighting functions to improve the resolution of the time delay estimate.
One particularly effective correlator uses only the phase spectrum, setting the modulus of the
cross-spectrum to unity over the frequency range of analysis. This is the so-called phase
transform (PHAT) correlator [4], which gives equal weighting to all frequencies, effectively
increasing the bandwidth over which the time delay is estimated, compensating to some extent
4
for the low-pass filtering properties of the pipe [7]. The pre-whitening approach can, however,
cause errors in the time delay estimate if there are additional phase shifts in the measurements
caused by the dynamic behaviour of the pipe system. Such phase shifts have been observed by
Gao et al. [6] in hydrophone measured data and by Almeida [8] in accelerometer measured
data. The associated increase in the modulus of the cross-spectrum at the frequencies where
there is an additional 90º of phase shift suggest that the dynamics is due to resonance behaviour.
The resonance behaviour is highly undesirable from the perspective of leak detection, as it can
significantly reduce the bandwidth over which the BCC extracts time delay information.
The aim of this paper is to describe a method to remove the additional phase shifts due to
the resonances in the system, so that a much wider bandwidth can be utilized, thus improving
the shape of the cross-correlation function, making it easier to determine the time delay
estimate. To achieve this, the ROTH correlator is used (as this is based on the frequency
response function (FRF) of the pipe system [4, 9]), which facilitates the visualization of the
resonance effects. Using a model of a resonator capable of capturing the additional phase shift
behaviour observed in measured data, these phase shifts are subsequently removed leaving only
the phase spectrum due to time delay. This process is illustrated using some experimental data
measured on an actual PVC water pipe.
2. Problem statement
Fig. 1 depicts a typical arrangement for water leak detection based on cross-correlation.
Pipe fittings such as meters, values and fire hydrants are used as access points for the
installation of the acoustic/vibration sensors such as hydrophones and accelerometers. The leak
generates broadband noise, which propagates along the pipe, and the difference in the arrival
times of the noise at the sensors (time delay) is used to determine the position of the leak. This
is given by [6],
5
1d 2d
d
1( )x t
2 ( )x t CORRELATOR Cross-correlation
function 1 2
( )x xR
Water pipe
Sensor 2 Sensor 1
Surface of the Ground
Leak
Fig. 1. Schematic of the process of leak detection in a buried water pipe using acoustic signals
with a leak bracketed by two sensors.
01
2
d cTd
(1)
where c is the propagation speed of the leak noise (wave speed); d is the distance between the
sensors; and 0 2 1( )T d d c is the time delay estimate. In many cases the wave speed is
estimated from tables, but it can also be measured in-situ [10]. The time delay T0 is estimated
from the peak in the cross-correlation function between the two measured signals 1( )x t and
2 ( )x t , which is given by [4]
1 2
1( ) ( )
2
j
x xR W e d
, (2)
where 1 2
( ) ( ) ( )x xW S , in which ( ) is given in Table 1 for the three correlators
considered in this paper; 1j ; 1 2
1 2 1 2
( )( ) ( ) x xj
x x x xS S e
is the cross spectral density
(CSD) function, in which 1 2
( )x xS and 1 2
( )x x are the modulus and phase spectra between
6
the signals 1( )x t and
2 ( )x t respectively; and is circular frequency. The time delay estimate
can also be determined from the frequency domain data by [5]
2
10 2
2
1
N
i i i
i
N
i i
i
W
T
W
, (3)
where the subscript i denotes the variable at the i-th frequency. Thus, the peak in the correlation
function corresponds to the gradient of a straight line which is fitted to the measured phase in
a weighted least squares sense.
Sensor 1
Sensor 2
Fig. 2. Plan view of the experimental leak detection facility at the NRC, Canada [11].
Measured signals from a pipe rig, specially constructed for water leak detection at the
National Research Council campus in Canada, are used to illustrate some issues. The
description of the test site and measurement procedures are given in [2, 11], and a plan of the
site is shown in Fig. 2. Noise from a leak, an illustration of which is shown in Fig. 3(a), was
measured using hydrophones, one of which is shown in Fig. 3(b). The distance between the
7
measurement points was 102.6 m, and the distance of the upstream measurement point from
the leak was 73.5 m. The signals of 66 s duration were each passed through an anti-aliasing
filter with the cut-off frequency set at 200 Hz and then digitized at a sampling frequency of
500 Hz.
(a) (b)
Fig. 3. Photographs illustrating a leak and a measurement point: (a) illustrative picture of a leak
before backfill; (b) one of the measurement positions [2].
(a) (b)
Frequency (Hz)
Po
wer
Sp
ectr
al d
ensi
ty (
dB
re
1P
a2/H
z)
Frequency range for analysis
Pow
er s
pec
tral
den
sity
(dB
re
1P
a2/H
z)
Frequency (Hz)
Cro
ss P
ow
erS
pec
tral
den
sity
(d
B r
e 1P
a2/H
z)
Frequency range for analysis
Cro
ss s
pec
tral
den
sity
(dB
re
1P
a2/H
z)
(c) (d)
Hydrophone
8
Frequency range for analysis
Frequency (Hz)
Co
her
ence
P
has
e (r
ad)
Frequency (Hz)
Frequency range for analysis
Fig. 4. Processed data from the test rig shown in Fig. 2: (a) PSDs of the two measured signals.
Red solid line, upstream position; blue dashed line, downstream position; (b) modulus of the
CSD; (c) coherence; (d) phase of the CSD. Solid blue line, measured data; red dashed line,
straight line approximation.
Some processed data from the two leak noise signals are shown in Fig. 4. These were
determined using a 1024-point FFT, a Hanning window (with 50% overlap) and spectral
averaging. Figure 4(a) shows the power spectral densities (PSDs) of the two measured signals.
Apparently, the PSD plots show that the signals attenuate with frequency and distance from the
leak. It is also clear that there is resonance behaviour, which is evident in each measurement.
The PSD with the resonant peak at 56 Hz is for the measurement downstream, which is closer
to the leak, and consequently has a higher value than the PSD with a peak at 83 Hz, which is
upstream of the leak. At frequencies greater than the resonance frequencies, the signal levels
attenuate even more rapidly with frequency, and this has an important effect on the behaviour
of the correlators. This is discussed in Section 4. The CSD between the two signals is shown
in Fig. 4(b). The two resonant peaks are evident in this plot as is the attenuation due to the low-
pass filtering properties of the pipe. The coherence is plotted in Fig. 4(c), which shows that the
bandwidth over which there is a reasonable linear relationship between the two signals is
9
between about 5 Hz and 90 Hz. Outside this bandwidth, the signals are contaminated with high
degree of uncorrelated noise. The bandwidth 5 - 90 Hz is thus chosen for subsequent analysis,
and this is shown on the graphs in Fig. 4. The phase of the CSD is shown in Fig. 4(d). The
measured data is plotted as a blue solid line, and overlaid on this is a straight line, which is
what the (ideal) phase would be if the pipe was infinite in length and located in an infinite
medium, so that there were no reflections or resonance behaviour. The ideal phase was
determined by multiplying the estimated time delay (see Section 4), the angular frequency and
minus one. The resonances at 56 Hz and 83 Hz are evident, where the phase first deviates from
the straight line by rads and then returns to the straight line respectively. Also evident are
the deviations from the phase at very low frequencies, less than about 20 Hz, which are due to
acoustic reflections in the pipe system [12].
The modulus of the CSD attenuates rapidly with frequency because of (a) the damping
properties of the pipe system, and (b) the resonances. As a result, the bandwidth of the signals
used for TDE is restricted to low frequencies. This is particularly relevant for the BCC function
in which the weighting function, 1 2 1 2
( ) ( )x x x xW S , which suffers significant attenuation at
high frequencies as can be seen in Fig. 4(b). If the PHAT correlator is used to overcome the
adverse effects of a small weighting function at high frequencies due to the damping in the pipe
system, then the phase shifts due to the resonances can cause an error in the TDE [8]. Thus,
there is the motivation to remove the resonance effects in the measured signals, and this is
further developed in the next section.
3. Process for improving the shape of the cross-correlation function
Conceptually, the passage of a leak noise signal from the leak to the sensor can be thought
of as passing through a pipe filter and then a resonator. The resonance effects can thus be
removed if the FRF for each resonator is known. If it is assumed that there is a resonator at
sensor 1 with FRF D1 and a resonator at sensor 2 with FRF D2, then provided that the
10
parameters of the resonators are known, the effects of the resonances can be removed from the
measured data by simply multiplying ( )W in Eq. (2) by D1/D2 for 1 2
( )x xS and by D2/ D1 for
2 1( )x xS , where the subscripts denote the resonators at sensors 1 and 2 respectively. The FRF
of a generic resonator is given by
2
2 2
2( )
2
n n
n n
jD
j
(4)
where n and are the natural frequency and damping ratio of the resonator respectively.
Table 1 The weighting functions used in the correlators
Correlator ( ) 1 2
( )x xW
BCC 1 1 2
( )x xS
PHAT 1 2
1
( )x xS 1
ROTH 1 1
1
( )x xS or
2 2
1
( )x xS 1 2
1 1
( )
( )
x x
x x
S
S
or
2 1
2 1
2 2
( )( )
( )
x x
x x
x x
SW
S
To determine the properties of the resonators, it is preferable to use the ROTH correlator,
as this is similar to a conventional measured FRF as can be seen in the final column of Table
1. This means the ROTH correlator will contain both resonances and anti-resonances if they
exist (which are the features related to the pipe dynamics that are of interest). Because there
are two measurement positions, there are two possible correlations using the ROTH processor,
which are from sensor 1 to sensor 2 and vice versa. Note that for the ROTH processor between
sensors 1 and 2, 2 1 1 2 1 1
( ) ( ) ( )x x x x x xW S S , and between sensors 2 and 1,
2 1 2 1 2 2( ) ( ) ( )x x x x x xW S S . The modulus and phase of the two ROTH processors are
calculated for the data from the test rig and are shown in Figs. 5(ai, aii) and (bi, bii) as solid
11
blue lines respectively. In Fig. 5(ai) it can be seen that a resonance appears in modulus of the
FRF between sensors 1 and 2, but in the FRF between sensors 2 and 1 an anti-resonance and a
resonance peaks are evident in Fig. 5(bi). The deviation of the phase from straight-line
behaviour associated with leak noise is also clear in both FRFs shown in Figs. 5(aii) and (bii).
Note that the only difference between the phase plots is one of sign.
(a) (bi)
Cro
ss P
ow
erS
pec
tral
den
sity
(d
B r
e 1P
a2/H
z)
Frequency (Hz)
Cro
ss s
pec
tral
den
sity
(dB
re
1P
a2/H
z)
Cro
ss s
pec
tral
den
sity
(dB
re
1P
a2/H
z)
Frequency (Hz)
(aii) (bii)
Frequency (Hz)
Phas
e (r
ad)
Phas
e (r
ad)
Frequency (Hz)
Fig. 5. Modulus and phase of the ROTH processor: (a) weighted CSD between sensors 1 and
12
2; (b) weighted CSD between sensors 2 and 1. (ai) and (bi) modulus; (aii) and (bii) phase. Solid
blue line, measured data; dashed red line, simulations using the model described in Appendix
A.
As mentioned previously, the parameters for the resonators need to be known if the dynamic
effects due to the resonance behaviour are to be removed from the data. These parameters
include the natural frequency and damping ratio. The natural frequencies can be determined
directly by inspection of the PSDs, and the damping can be determined by curve fitting
measured data to a model of the ROTH correlator at frequencies close to the resonance
frequencies. Such a model, including the effects of the resonators, is given in Appendix A.
Setting the natural frequencies of the resonators for sensor 1 and sensor 2 to be 56 Hz and 83
Hz respectively, the damping was found to be 0.022 for each resonator. The model results
are overlaid with the measured data in Fig. 5. It can be seen that the model captures the modulus
of the ROTH correlator between sensors 2 and 1, in Fig. 5(bi), but it does not fare as well for
the measurement between sensors 1 and 2, in Fig. 5(ai). The reason for this is that the signals
are dominated by high noise levels in the PSD at Position 1 above about 90 Hz. The predicted
modulus is much larger at higher frequencies in Fig 5(ai) than in Fig 5(bi), and is not captured
in the measurements because of the high level of incoherent noise. Note that the model captures
the behaviour of the phase well in both cases, as can be seen in Figs. 5(aii) and (bii).
Once the parameters of the resonators have been determined, the effects of the resonances
can be removed from the measured data by simply multiplying ( )W in Eq. (2) by D1/D2 for
1 2( )x xS and by D2/D1 for
2 1( )x xS with respect to the two ROTH processors. The resulting
CSDs for the ROTH correlators are shown in Fig. 6 as a dashed blue line. Also in this figure
are the measured data shown in Fig. 5. It is clear from Figs. 6(ai) and (bi) that the peak and the
trough are “flattened” in the moduli, giving smoother curves. Also, more importantly, the
deviations from the phase due to the resonance effects are removed, as clearly evident in Figs.
13
6(aii) and (bii).
(ai) (bi)
Cro
ss P
ow
erS
pec
tral
den
sity
(d
B r
e 1
Pa
2/H
z)
Frequency (Hz)
Cro
ss s
pec
tral
den
sity
(d
B r
e 1
Pa2
/Hz)
Frequency (Hz)
Cro
ss s
pec
tral
den
sity
(d
B r
e 1
Pa2
/Hz)
(aii) (bii)
Frequency (Hz)
Ph
ase
(rad
)
Frequency (Hz)
Ph
ase
(rad
)
Fig. 6. Modulus and phase of the ROTH processor: (a) weighted CSD between sensors 1 and
2; (b) weighted CSD between sensors 2 and 1. (ai) and (bi) modulus; (aii) and (bii) phase. Solid
red line, raw measured data; dashed blue line, data after the resonance effects have been
removed.
14
4. Cross-correlation functions for TDE
The correlation results for the BCC, PHAT and ROTH correlators are shown in Fig. 7(ai-
aiv) for the raw data, and in Fig. 7(bi-biv) for the data processed so that resonance effects are
removed. The dashed vertical line shows the true time delay of 92 ms, assuming a wave speed
of 482 m/s calculated based on the measurements from an in-bracket leak source [2]. Consider
first the BCC correlator without and with processing, which is respectively shown in Figs. 7(ai)
and (bi). The peak related to the time delay in both the figures is 94 ms, a percentage difference
of about 2%. The additional peaks in both figures correspond to the reflections at the ends of
the pipes at the measurement positions. These reflections are important at low frequencies
because of the small attenuation of waves at these frequencies, and this phenomenon is
discussed in detail in [12]. They manifest themselves strongly in the BCC function because of
the spectral characteristics of the modulus of the CSD shown in Fig. 4(b), which weights low
frequencies much more than high frequencies. The resonance effects do not have a large effect
on the BCC function, but can be seen as additional ripples in Fig. 7(ai) compared with Fig.
7(bi).
(ai) (bi)
(aii) (bii)
No
rmal
ised
Cro
ss-c
orr
elat
ion f
un
ctio
n
15
(aiii) (biii)
(aiv) (biv)
No
rmal
ised
Cro
ss-c
orr
elat
ion f
un
ctio
nN
orm
alis
edC
ross
-co
rrel
atio
n f
un
ctio
nN
orm
alis
edC
ross
-co
rrel
atio
n f
un
ctio
n
Time delay (s) Time delay (s)
16
Fig. 7. Cross-correlation functions: (a) with the resonance effects present; (b) with the
resonance effects removed. (ai) and (bi) BCC; (aii) and (bii) PHAT; (aiii) and (biii) ROTH
between sensors 1 and 2; (aiv) and (biv) ROTH between sensors 2 and 1. The dashed vertical
line corresponds to the true time delay of 92 ms.
Consider now the PHAT correlation results without and with processing, which is
respectively shown in Figs. 7(aii) and (bii). It is clear that the PHAT correlator does not give
the correct time delay if the bandwidth 5-90 Hz is chosen as shown in Fig. 7(aii), but it does
give a good estimate of 90 ms if the resonance effects are removed as can be seen in Fig. 7(bii).
Furthermore, as shown in Fig. 7(bii), the PHAT correlator also suppresses the additional peaks
due to the reflections, removing some ambiguity in the interpretation of a cross-correlation
function. The reason for this is that it gives equal weighting to low and high frequency data
within the bandwidth chosen for analysis. Note that PHAT correlator can give a good estimate
of the time delay without the removal of the resonance effects, but this can only be achieved
when the bandwidth over which the analysis is conducted is restricted to low frequencies,
below the first resonance [8].
For completeness, the ROTH correlation function between sensors 1 and 2 is shown in Figs.
7(aiii) and (biii). The correlation function without removing the resonance effects is shown in
Fig. 7(aiii). It can be seen that this appears to be an impulse response with time advance. The
peak in this does not correspond to the time delay estimate. Removing the resonance effects
effectively reduces the slowly decaying oscillations from the correlation function, leaving only
a sharp peak that occurs at 90 ms. Examining the modulus of the CSD for the ROTH correlator
in Fig. 6(ai), shows that this increases with frequency up to about 80 Hz, and thus weights the
higher frequencies more than lower frequencies, which results in the sharp peak. Broadly
similar results are seen for the ROTH correlation function between sensors 2 and 1 as shown
in Figs. 7(aiv) and (biv). The differences between the ROTH correlation functions with the
17
resonance effects removed, which can be seen in Figs. 7(biii) and (biv), can be attributed to the
respective moduli of the CSDs shown in Figs. 6(ai) and (bi). It can be seen that the weighting
functions are different, with the ROTH correlation function between sensors 1 and 2 weighting
the higher frequencies more that the ROTH correlation function between sensors 2 and 1.
(a)
Frequency range for analysis
Wei
gh
tin
g f
un
ctio
n (
dB
)
Frequency (Hz)
(b)
Wei
gh
tin
g f
un
ctio
n (
dB
)
Frequency (Hz)
18
Fig. 8. Weighting functions ( )W : (a) without the resonance effects removed; (b) with
resonance effects removed. Solid green line, BCC; straight solid line at 0 dB, PHAT; dashed-
dotted blue line, ROTH between sensors 1 and 2; dashed blue line, ROTH between sensors 2
and 1.
To aid interpretation into why the correlators have different cross-correlation function
shapes, the weighting functions ( )W for the four correlators considered (the BCC, PHAT
and ROTH (×2)) are shown for the raw data in Fig. 8(a) and for the data with the resonance
effects removed in Fig. 8(b). Note that the weighting functions for the raw data are simply the
moduli of the cross-spectra. Moreover, the phases are the same for all three correlators, and are
shown in Fig. 6(aii) or 6(bii). Examining Figs. 8(a) and 8(b), it is clear why the BCC is not
greatly affected when the resonances are improved. It is also clear why the ROTH processers
benefit by the removal of the resonances as the weighting functions become “flatter” with
frequency and so give broadly similar results to the PHAT correlator. However, it should be
noted that there is a choice of which ROTH correlator to choose from, and depending on
particular situations, it is possible that they will give different time delay estimates.
5. Discussion
By examining the results shown in the previous section, it is clear that with some simple
processing of the data, it is possible to significantly improve the shape of the cross-correlation
function, and hence aid interpretation of this in the determination of time delay and location of
a leak. This does, however, involve a certain amount of expertise, in which the analyst is able
to observe resonance behaviour in the data. If this is possible then their effects can be removed
using the procedure described in this paper (either one or more resonances), and the bandwidth
over which the data can be used for TDE can be increased. Further, this allows the PHAT
correlator to be used with confidence, over the bandwidth defined by good coherence, to
remove the damping effects of the pipe (wave attenuation at high frequencies) and to suppress
19
peaks due to reflections, to give a clear and accurate estimate of the time delay as shown in Fig.
7(bii). It remains an open problem on how to determine the resonance effects in leak noise data
automatically and remove them without human intervention.
6. Conclusions
In this paper, a way to improve the shape of the cross-correlation function for leak detection
in plastic water distribution pipes has been investigated. Using experimental data, it has been
shown that significant improvement can be achieved by determining the FRFs for any
resonators responsible for peaks in the CSD functions, and then processing the data using the
FRFs of these resonators. The parameters of the resonators are best determined by examining
the ROTH correlator as this is similar to the FRF of the system, and thus facilitates the
identification of both resonances and anti-resonances if they exist. The net effect of processing
the data is to remove any deviations of the phase from that expected for a pure time delay and
to smooth the modulus of the CSD. Using the modified phase spectrum, the PHAT correlator
can be used over the frequency bandwidth in which there is good coherence between the two
measured signals, resulting in an unambiguous and clear estimate of the time delay. There are
thus significant advantages in applying the process described in this paper. The current
limitation, however, is that the procedure has to be applied manually, which will restrict its
application in practical acoustic correlators. The challenge now is to develop an algorithm to
carry out the procedure automatically.
Acknowledgements
The authors gratefully acknowledge the financial support of the CAS Hundred Talents
Programme and FAPESP, project No. 2013/50412-3, and Osama Hunaidi from the National
Research Council of Canada who provided the test data.
20
References
[1] Fuchs HV, Riehle R. Ten years of experience with leak detection by acoustic signal
analysis. Appl Acoust 1991; 33: 1–19.
[2] Hunaidi O, Chu WT. Acoustical characteristics of leak signals in plastic water
distribution pipes. Appl Acoust 1999; 58: 235-254.
[3] Bendat JS, Piersol AG. Engineering of applications of correlation and spectral analysis.
2nd ed. New York: Wiley; 1993.
[4] Gao Y, Brennan MJ, Joseph PF. A comparison of time delay estimators for the detection
of leak noise signals in plastic water distribution pipes. J Sound Vib 2006; 292:552-570.
[5] Brennan MJ, Gao Y, Joseph PF. On the relationship between time and frequency domain
methods in time delay estimation for leak detection in water distribution pipes. J Sound
Vib 2007; 304: 213-223.
[6] Gao Y, Brennan MJ, Joseph PF, Muggleton JM, Hunaidi O. A model of the correlation
function of leak noise in buried plastic pipes. J Sound Vib 2004; 277: 133-148.
[7] Almeida FCL, Brennan MJ, Joseph PF, Whitfield S, Dray S, Paschoalini A. On the
acoustic filtering of the pipe and sensor in a buried plastic water pipe and its effect on
leak detection: an experimental investigation. Sensors 2014; 14: 5595-5610.
[8] Almeida FCL. Improved acoustic methods for leak detection in buried plastic water
distribution pipes. PhD Thesis, University of Southampton, 2013.
[9] Roth PR. Effective measurements using digital signal analysis. IEEE Spectrum 1971; 8:
62-70.
[10] Almeida FCL, Brennan MJ, Joseph PF, Dray S, Whitfield S, Paschoalini AT. Towards an
in-situ measurement of wave velocity in buried plastic water distribution pipes for the
purposes of leak location. J Sound Vib 2015; 359: 40-55.
21
[11] Hunaidi O, Chu W, Wang A, Guan W. Detecting leaks in plastic pipes, J Am Water
Works Ass 2000; 92: 82–94.
[12] Gao Y, Brennan MJ, Joseph PF. On the effects of reflections on time delay estimation for
leak detection in buried plastic water pipes. J Sound Vib 2009; 325(3): 649-663.
[13] Pinnington RJ, Briscoe AR, Externally applied sensor for axisymmetrical waves in a
fluid-filled pipe. J Sound Vib 1994; 173: 503-516.
[14] Gao Y, Sui F, Muggleton JM, Yang J, Simplified dispersion relationships for fluid-
dominated axisymmetric wave motion in buried fluid-filled pipes. J Sound Vib 2016;
375: 386-402.
[15] Gao Y, Liu Y, Muggleton JM. Axisymmetric fluid-dominated wave in fluid-filled plastic
pipes: loading effects of surrounding elastic medium. Appl Acoust 2017; 116: 43–49.
Appendix A: Model for the ROTH correlator
In this appendix, a model of the ROTH correlator is described. First, however, a model of
leak noise propagation in an infinite pipe (with no resonators), is described. As the pipe is
assumed to be of infinite length, there are no wave reflections from boundaries. The leak
generates noise that propagates as a non-dispersive coupled plane wave involving the water
and the pipe-wall away from the leak to the measurement positions [6, 13]. The pressure of this
wave at position u is given by [6]
( , ) ( ) ,p u P H u (A1)
where ( )P is the amplitude of the acoustic pressure at the leak location, and
i, u u cH u e e is the FRF between the pressure at the leak position and the pressure at
the measurement position, in which u is the distance from the leak, β is an attenuation factor
and c is the speed of wave propagation. For the pipe discussed in this paper c is taken to be 482
m/s and β is taken to be 41.58 10 s/m. Note that these are taken to be constants. Recent work
22
by Gao et al. [14, 15] has shown that for plastic water pipes, the presence of the surrounding
soil increases both the propagation wavespeed and attenuation with frequency compared to the
in- vacuo case. However, they may be assumed to be constant for the purposes of this paper,
because in the experimental work leak noise signals are confined to low frequencies.
The CSD between the two measured signals is given by [6]
1 2 1 2( ) ( , ) ( , )x x llS S H d H d (A2)
where * denotes the complex conjugate; llS is the PSD of the leak, which is assumed to be
constant in the frequency range of interest; and the distances refer to those given in Fig. 1. The
PSDs of the leak signals are given by
1 1
2
1( ) ( , )x x llS S H d ; 2 2
2
2( ) ( , )x x llS S H d (A3a, b)
respectively. Noting that 1 2
( ) ( ) ( )x xW S from Section 2, where ( ) is given in Table
1, the CSD for the ROTH correlator between sensors and 1 and 2 and sensors 2 and 1 are
respectively given by
1 2 0
1 2( )
d d j T
x xW e e
; 1 2 0
2 1( )
d d j T
x xW e e
(A4a, b)
To include the effects of the resonances for the ROTH correlator between sensors 1 and 2,
1 2( )x xW is simply multiplied by D2/D1 where Di is the FRF of a resonator given by Eq. (4),
and the subscripts denote sensors 1 and 2 respectively. To include the effects of the resonances
for the ROTH correlator between sensors 2 and 1, 2 1
( )x xW is multiplied by D1/D2.